Pith. sign in

REVIEW 3 major objections 5 minor 58 references

Lithium-Projected Phonon Spectral Distributions as Robust Descriptors of Ionic Conductivity in Solid Electrolytes

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The complete lithium-projected phonon spectrum, viewed through Wasserstein geometry, carries reproducible information about experimental room-temperature ionic conductivity that scalar softness measures and total phonon spectra miss.

desk verdict A carefully designed comparative study showing full Li-PDOS beats total-DOS and static baselines on OBELiX, but the Li-projection validation rests on only 10 references and the mechanistic gloss is fragile — worth a serious referee, not a desk reject. read the letter →

arxiv 2607.20591 v1 pith:XTPOLT2K submitted 2026-07-22 cond-mat.mtrl-sci

Gajendra Bohara , Ramakrishna Podila This is my paper · ORCID
classification cond-mat.mtrl-sci
keywords solidelectrolyteionicconductivitylithium-projectedphonondensityofstatesWassersteinkernelsoftnessmachine-learninginteratomicpotentialmaterialsinformaticslatticedynamics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tests whether the full frequency-resolved vibrational spectrum of lithium ions—not just a single 'softness' number—can predict measured room-temperature ionic conductivity across chemically diverse solid electrolytes. Using machine-learned forces to generate harmonic spectra for hundreds of crystallographically resolved materials, it finds that higher-conductivity materials systematically shift normalized lithium spectral weight toward low frequencies, a pattern that replicates in an untouched test set. A Wasserstein kernel on the lithium-projected density of states yields held-out R²=0.444, versus 0.012 for the total phonon spectrum and 0.181 for a static composition–structure baseline, and remains stable under stricter composition auditing. The paper argues that mobile-ion-resolved spectral distributions are robust comparative screening descriptors, while cautioning that they are not a universal causal law.

What carries the argument

The central object is the unit-area-normalized lithium-projected phonon density of states, treated as a probability distribution over frequency. The Wasserstein-1 distance between such distributions, implemented through quantile embeddings, provides a kernel that respects the ordering of the frequency axis, unlike bin-wise vector representations. The paper also uses cumulative-frequency descriptors such as the fraction of spectral weight below 2 and 5 THz and the 5% quantile frequency. These carry the argument by converting raw spectra into distribution-aware features that separate conductive from non-conductive materials.

What would settle it

A direct test would be to recompute lithium-projected phonon spectra for a few dozen representative materials using first-principles density-functional-theory forces and check whether the low-frequency lithium weight redistribution and the Wasserstein-model R² survive; if the correlation with conductivity collapses under the more accurate spectra, the descriptor is an artifact of the approximate force model.

Watch

Extended reading notes

Core claim

The central claim is that the complete lithium-projected phonon density of states (Li-PDOS)—the distribution of vibrational modes weighted by lithium motion—contains transferable information about experimental ionic conductivity that is lost when the spectrum is compressed to a band center or other scalar, and also lost when only the total density of states is used. In the paper's primary cohort of 260 materials, the Li-PDOS Wasserstein model achieves held-out R²=0.444 versus 0.012 for total DOS and 0.181 for a static kernel. The low-frequency redistribution direction and cumulative-onset shift replicate in the untouched test set, and the R² remains near 0.45 under strict and exact-compositi

Load-bearing premise

The calculated lithium phonon spectra, generated from a machine-learned interatomic potential applied to simplified ordered versions of the real crystal structures, accurately represent the true lattice vibrations of these diverse electrolytes.

Editorial extensions

If this is right

  • Screening workflows should retain mobile-ion-projected spectra instead of only total DOS or scalar softness descriptors.
  • A Wasserstein-kernel model on Li-PDOS can rank candidate electrolytes before expensive molecular dynamics or first-principles diffusion calculations.
  • The observed low-frequency redistribution provides a specific, testable spectral signature (increased Li weight below ~5 THz, earlier cumulative onset) for prioritizing materials.
  • Audit and validation practices—checking composition, integerization, and comparing against first-principles references—become standard parts of MLIP-based phonon screening.
  • Total-DOS agreement with reference calculations cannot validate a mobile-ion projection; species-resolved validation is necessary.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the Li-PDOS Wasserstein signal is reproducible across datasets beyond the one used here, it could be combined with anharmonic and defect-aware simulations to separate genuine lattice-dynamical softness from compositional confounding.
  • The family-attenuation of scalar descriptors suggests the full-distribution model may work by capturing cross-family differences; a natural extension is to test whether the Wasserstein kernel still adds value for substitutions within a single electrolyte family.
  • Because the harmonic spectrum is computed at zero temperature, the descriptor may be a proxy for something else—such as coordination or polarity—so a decisive test would be to correlate the same low-frequency redistribution with independently measured migration barriers or ab initio molecular-dynamics conductances.
  • The systematic lithium-projection softening reported for the approximate force model suggests that calibrating against first-principles spectra for a small representative set could improve the descriptor's quantitative accuracy without losing its ranking power.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper tests whether the complete lithium-projected phonon density of states (Li-PDOS), treated as a distribution and compared via Wasserstein geometry, is a robust descriptor of experimental room-temperature ionic conductivity. Using MatterSim MLIP forces and Phonopy, the authors compute harmonic total and Li-projected spectra for OBELiX entries after an audit of composition/structural fidelity, preserving the official train/test split. They construct nested cohorts (primary 260, strict 241, exact 168) and report that higher conductivity is associated with low-frequency redistribution of normalized Li spectral weight. In the untouched test set, a Li-PDOS Wasserstein kernel achieves R²=0.444, versus 0.012 for total DOS and 0.181 for a static composition–structure kernel, with similar R² values in the strict and exact cohorts. The paper includes extensive sensitivity analyses, multiplicity control, external DFT reference comparisons, and acknowledges limitations related to MLIP softening and family confounding.

Significance. If the claims hold, the paper provides a practical screening descriptor for solid electrolytes and a methodologically careful template for evaluating distribution-valued descriptors. The validation-first design is a clear strength: the official split is respected, frequency-window selection is training-only, permutation/FDR corrections are applied, and the code/data are released. External comparison with NIMS/PhononDB, though limited in size, is a valuable addition. The authors are appropriately cautious about causal interpretation, and the paper's explicit discussion of MLIP softening and family attenuation is honest. However, the central claim depends on the Li projection being a faithful proxy for true lattice dynamics, and the current validation leaves a nontrivial risk of conductivity-correlated bias that could explain part of the Li-PDOS advantage.

major comments (3)
  1. [§2.4, §3.2, Fig. 1] The external validation is too limited to rule out a conductivity-correlated bias in the Li projection. The mean W1 distance is 0.731 THz and the mean centroid error is -0.711 THz for Li-PDOS, based on only 10 phase-matched materials (20 comparisons counting total and Li separately). This is a systematic, projection-dependent softening. If the MatterSim Li-force error correlates with composition or conductivity—e.g., because fast-ion conductors are more anharmonic or contain heavier elements—the held-out R²=0.444 advantage over total DOS could reflect a computational artifact rather than physical information. The paper itself notes that "total-DOS agreement cannot validate a species projection" but does not provide a test of whether the Li projection error is conditionally independent of the target. Please expand the reference set to a broader, conductivity-stratified sample, or demonstr
  2. [§3.5, Fig. 4] The family-confounding analysis is not extended to the prediction comparison. The marginal training correlation for Li fraction below 5 THz is ρ=0.538, but after family demeaning it drops to approximately 0.150, and the test interval crosses zero. Moreover, the Li-vs-total HSIC ordering reverses in the exact cohort (Li 0.125 vs total 0.184). These observations show that the pooled Li-PDOS advantage may be driven by family composition. The paper reports residual HSIC in some cohorts, but not a family-demeaned or within-family prediction comparison between Li and total spectra. To support the claim that Li-PDOS contains information "not captured by total DOS," please report leave-one-family-out or family-demeaned R² for the Li-Wasserstein vs total-Wasserstein and static models on the official test set, or otherwise quantify how much of the Li advantage survives control for family identity.
  3. [§3.5, Fig. 4c] The p-values in Figure 4c are presented without clearly distinguishing which permutation null (global vs within-family) each p-value refers to. The text states HSIC is significant "under both global and within-family permutations," but the figure symbol legend and caption imply all listed p-values come from the same permutation procedure. Please clarify by explicitly labeling the permutation type for each p-value, or moving the within-family p-values to a separate panel/table.
minor comments (5)
  1. [§3.6] Typo: "rooot-mean squared" should be "root-mean squared" in the first paragraph.
  2. [§3.4 heading] Heading contains a stray space: "F requency-Resolved" should be "Frequency-Resolved."
  3. [§2.2] The software is referred to as both "PhonoPy" and "Phonopy" in the text; please standardize to the official spelling "Phonopy."
  4. [§2.4] The description "20 independently generated phonon-calculation database comparisons" is somewhat misleading because the 20 comparisons come from 10 materials with two representations each. The text later clarifies this, but readers may initially interpret it as 20 independent crystal structures.
  5. [§3.5] The exact-cohort HSIC reversal is discussed in the text but not shown quantitatively in Figure 4; please point more explicitly to the supporting figure or table showing the exact-cohort HSIC values and the permutation intervals.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity found: the held-out prediction is a genuine out-of-sample test and the descriptor is computed independently of the conductivity labels.

full rationale

The paper's derivation chain is not circular. The central prediction uses Li-PDOS distributions computed from MatterSim-Phonopy, which are generated from crystal structures and force constants without any use of the experimental conductivity labels. The Wasserstein kernel is a fixed geometric representation of these distributions; its hyperparameters are selected on the training set only, and the official OBELiX test split is untouched until final evaluation. Frequency-window associations are likewise chosen on training data and then replicated directionally on the test set, so they are genuine out-of-sample confirmations rather than fitted inputs renamed as predictions. The static and total-DOS baselines are evaluated on the same cohorts, and the reported R² values are held-out metrics, not training fits. The external validation against NIMS/PhononDB is independent evidence, even if limited in size; a small validation set or a possible MLIP bias is a correctness or generalizability concern, not a circularity concern. No equation in the paper defines conductivity from a fitted parameter, and no load-bearing claim reduces to a self-citation or an imported uniqueness theorem. The paper also explicitly acknowledges limitations, including family attenuation and the reversal of Li-versus-total HSIC ordering in the exact cohort, which further indicates that the analysis is not structured to force the stated conclusion.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on domain assumptions about MLIP phonon accuracy, harmonic approximation, and structural representation of disordered materials, plus two data-fitted model-hyperparameter entries. There are no invented physical entities, new forces, or new particles. The authors explicitly constrain the mechanistic interpretation, which lowers the ad-hoc content.

free parameters (2)
  • KRR regularization and kernel bandwidth = not reported; selected via grouped nested cross-validation on official training set
    These are model hyperparameters chosen from training data only; they influence absolute R² but are not fitted to test labels. They are the main data-fitted numbers in the descriptor pipeline.
  • Training quartile thresholds for spectral stratification = quartiles of log conductivity in the primary training cohort
    Used only to construct descriptive high-minus-low redistribution curves; the predictive model and test replication do not depend on these thresholds.
assumptions (4)
  • domain assumption Harmonic phonon DOS from finite-displacement force constants adequately represents the vibrational information relevant to conductivity screening.
    Invoked throughout §2.2-2.3 and §3.2; the paper uses harmonic Phonopy spectra and does not include anharmonicity or finite-temperature mode occupancy.
  • domain assumption MatterSim universal MLIP forces approximate DFT sufficiently across the 260-material cohort.
    Invoked in §2.2 and bounded in §3.2 using only ten phase-matched PhononDB comparisons; systematic Li-projection softening remains.
  • domain assumption Ordered or integerized supercell approximations of disordered battery materials preserve the relevant lithium sublattice and projected phonon modes.
    Introduced in §2.1 and audited in §3.1 via PASS/INTEGERIZED/WARNING categories; the exact-cohort analysis limits but cannot eliminate the risk.
  • domain assumption OBELiX curated room-temperature experimental conductivities, despite heterogeneous measurement protocols, are comparable under log transformation.
    Used in §2.1 as regression targets; the paper treats them as given and only tests censoring conventions, not inter-laboratory comparability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lithium-Projected Phonon Spectral Distributions as Robust Descriptors of Ionic Conductivity in Solid Electrolytes." pith.science (2026). https://pith.science/paper/XTPOLT2K

@misc{pith2026260720591,
  author       = {Pith},
  title        = {Pith review of: Lithium-Projected Phonon Spectral Distributions as Robust Descriptors of Ionic Conductivity in Solid Electrolytes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XTPOLT2K}},
  note         = {Machine review of arXiv:2607.20591}
}
abstract

Lattice dynamics are widely invoked in the design of solid electrolytes, yet phonon information is commonly compressed into a band center or another scalar softness measure. Here we test whether the complete lithium-projected phonon density of states (Li-PDOS) provides a reproducible descriptor of experimental room-temperature ionic conductivity. MatterSim forces and Phonopy were used to generate harmonic total and Li-projected spectra for crystallographically resolved entries in the OBELiX dataset. A composition and structure audit defined a primary cohort of 260 materials (212 train and 48 test), a strict cohort of 241, and an exact-composition cohort of 168. Across 20 independently generated phonon-calculation database comparisons, the mean Wasserstein-1 distance was 0.542 THz for total DOS and 0.731 THz for Li-PDOS, revealing broad agreement but systematic, projection-dependent softening. Higher conductivity was associated with redistribution of normalized Li spectral weight toward low frequencies: in the untouched test set, the Li fractions below 2 and 5 THz had Spearman coefficients of 0.333 and 0.374, while the 5$\%$ cumulative-frequency quantile had a coefficient of $-0.393$. A Wasserstein kernel on the full Li-PDOS achieved held-out $R^2=0.444$, compared with 0.012 for total DOS and 0.181 for a static composition--structure kernel. The Li model remained stable in the strict ($R^2=0.462$) and exact ($R^2=0.451$) cohorts. Family adjustment attenuated scalar associations, and Li-versus-total whole-spectrum dependence was cohort sensitive. The results therefore support mobile-ion-resolved spectral distributions as useful comparative screening descriptors, not as a universal causal softness law.

Figures

Figures reproduced from arXiv: 2607.20591 by the authors.

Figure 1
Figure 1. External validation against strict-phase NIMS/PhononDB references. (a) [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Conductivity-stratified spectral redistribution in the primary training cohort. (a) [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Frequency-resolved Spearman associations with experimental log conductivity. (a,b) [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Scalar and whole-spectrum dependence. (a) Train/test replication of selected Li [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Held-out prediction and audit-cohort sensitivity. (a) Official-test MAE for static, [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references

  1. [1]

    Manthiram, Arumugam and Yu, Xingwen and Wang, Shaofei , title =. Nat. Rev. Mater. , year =

  2. [2]

    , title =

    Janek, Jürgen and Zeier, Wolfgang G. , title =. Nat. Energy , year =

  3. [3]

    and Muy, Sokseiha and Grimaud, Alexis and Chang, Hao-Hsun and Pour, Nima and Lux, Simon F

    Bachman, John C. and Muy, Sokseiha and Grimaud, Alexis and Chang, Hao-Hsun and Pour, Nima and Lux, Simon F. and Paschos, Odysseas and Maglia, Filippo and Lupart, Saskia and Lamp, Peter and Giordano, Livia and Shao-Horn, Yang , title =. Chem. Rev. , year =

  4. [4]

    and Islam, M

    Famprikis, Theodosios and Canepa, Pieremanuele and Dawson, James A. and Islam, M. Saiful and Masquelier, Christian , title =. Nat. Mater. , year =

  5. [5]

    Albertus, Paul and Babinec, Susan and Litzelman, Scott and Newman, Aron , title =. Nat. Energy , year =

  6. [6]

    and Kötz, Otto and Koerver, Raimund and Braun, Patrick and Weber, André and Ivers-Tiffée, Ellen and Adermann, Tobias and Kulisch, Jörg and Zeier, Wolfgang G

    Randau, Simon and Weber, David A. and Kötz, Otto and Koerver, Raimund and Braun, Patrick and Weber, André and Ivers-Tiffée, Ellen and Adermann, Tobias and Kulisch, Jörg and Zeier, Wolfgang G. and Richter, Franz H. and Janek, Jürgen , title =. Nat. Energy , year =

  7. [7]

    and Meng, Ying Shirley , title =

    Banerjee, Abhik and Wang, Xuefeng and Fang, Chengcheng and Wu, Erik A. and Meng, Ying Shirley , title =. Chem. Rev. , year =

  8. [8]

    and Ceder, Gerbrand , title =

    Xiao, Yanhao and Wang, Yan and Bo, Shou-Hang and Kim, Jae Chul and Miara, Lincoln J. and Ceder, Gerbrand , title =. Nat. Rev. Mater. , year =

Show all 58 references
  1. [9]

    , title =

    Zhao, Qian and Stalin, Swetha and Zhao, Cheng-Zheng and Archer, Lynden A. , title =. Nat. Rev. Mater. , year =

  2. [10]

    Kerman, Kian and Luntz, Alan and Viswanathan, Viswanathan and Chiang, Yet-Ming and Chen, Zhongwei , title =. J. Electrochem. Soc. , year =

  3. [11]

    and Rettenwander, Daniel and Frömling, Till and Thaman, Henry L

    Porz, Lukas and Swamy, Tirumala and Sheldon, Brian W. and Rettenwander, Daniel and Frömling, Till and Thaman, Henry L. and Berendts, Stefan and Uecker, Reinhard and Carter, W. Craig and Chiang, Yet-Ming , title =. Adv. Energy Mater. , year =

  4. [12]

    and Yue, Jia and Fan, Xueliang and Wang, Fei and Chi, Ming and Leonard, Dean N

    Han, Fudong and Westover, Andrew S. and Yue, Jia and Fan, Xueliang and Wang, Fei and Chi, Ming and Leonard, Dean N. and Dudney, Nancy J. and Wang, Howard and Wang, Chunsheng , title =. Nat. Energy , year =

  5. [13]

    and Kim, Jae Chul and Mo, Yifei and Ceder, Gerbrand , title =

    Wang, Yan and Richards, William Davidson and Ong, Shyue Ping and Miara, Lincoln J. and Kim, Jae Chul and Mo, Yifei and Ceder, Gerbrand , title =. Nat. Mater. , year =

  6. [14]

    and Giordano, Livia and Chang, Hao-Hsun and Abernathy, Douglas L

    Muy, Sokseiha and Bachman, John C. and Giordano, Livia and Chang, Hao-Hsun and Abernathy, Douglas L. and Bansal, Dipanshu and Delaire, Olivier and Hori, Satoshi and Kanno, Ryoji and Maglia, Filippo and Lupart, Saskia and Lamp, Peter and Shao-Horn, Yang , title =. Energy Enviro...

  7. [15]

    and Zeier, Wolfgang G

    Krauskopf, Thorben and Pompe, Constantin and Kraft, Marvin A. and Zeier, Wolfgang G. , title =. Chem. Mater. , year =

  8. [16]

    and Ohno, Saneyuki and Delaire, Olivier and Shao-Horn, Yang and Zeier, Wolfgang G

    Krauskopf, Thorben and Muy, Sokseiha and Culver, Sean P. and Ohno, Saneyuki and Delaire, Olivier and Shao-Horn, Yang and Zeier, Wolfgang G. , title =. J. Am. Chem. Soc. , year =

  9. [17]

    and Maglia, Filippo and Lamp, Peter and Zeier, Wolfgang G

    Muy, Sokseiha and Voss, Johannes and Schlem, Roman and Koerver, Raimund and Sedlmaier, Stefan J. and Maglia, Filippo and Lamp, Peter and Zeier, Wolfgang G. and Shao-Horn, Yang , title =. iScience , year =

  10. [18]

    , title =

    Muy, Sokseiha and Schlem, Roman and Shao-Horn, Yang and Zeier, Wolfgang G. , title =. Adv. Energy Mater. , year =

  11. [19]

    and Chu, Dewei and Cazorla, Claudio , title =

    Sagotra, Arun K. and Chu, Dewei and Cazorla, Claudio , title =. Phys. Rev. Mater. , year =

  12. [20]

    and Shao-Horn, Yang and Henry, Asegun , title =

    Gordiz, Kiarash and Muy, Sokseiha and Zeier, Wolfgang G. and Shao-Horn, Yang and Henry, Asegun , title =. Cell Rep. Phys. Sci. , year =

  13. [21]

    Xu, Zhenming and Chen, Xi and Zhu, Hong and Li, Xin , title =. Adv. Mater. , year =

  14. [22]

    Song, Tao and Lin, Yuxiao and Wang, Da and Chen, Qianli and Ling, Chen and Shi, Siqi , title =. Adv. Energy Mater. , year =

  15. [23]

    ACS Appl

    Ouyang, Runxin and Yang, Yu and Guan, Chaohong and Zhu, Hong , title =. ACS Appl. Mater. Interfaces , year =

  16. [24]

    and Rosenbach, Carolin and Lin, Hung-Min and Osti, Naresh C

    Ding, Jingxuan and Gupta, Mayanak K. and Rosenbach, Carolin and Lin, Hung-Min and Osti, Naresh C. and Abernathy, Douglas L. and Zeier, Wolfgang G. and Delaire, Olivier , title =. Nat. Phys. , year =

  17. [25]

    and Gordiz, Kiarash and Spear, Nathaniel A

    Pham, Kim H. and Gordiz, Kiarash and Spear, Nathaniel A. and Lin, Andrew K. and Michelsen, Jonathan M. and Liu, Hanzhe and Vivona, Daniele and Blake, Geoffrey A. and Shao-Horn, Yang and Henry, Asegun and See, Kimberly A. and Cushing, Scott K. , title =. Mater. Horiz. , year =

  18. [26]

    , title =

    Li, Xinyu and Benedek, Nicole A. , title =. Chem. Mater. , year =

  19. [27]

    and Siegel, Donald J

    Greene, Samuel M. and Siegel, Donald J. , title =. Chem. Mater. , year =

  20. [28]

    Kim, Jiyeon and Lee, Donggeon and Lee, Dongwoo and Li, Xin and Lee, Yea-Lee and Kim, Sooran , title =. J. Phys. Chem. Lett. , year =

  21. [29]

    Putu and Oh, Jeong Moo and Kang, Jee-Hyun and Mun, Junyoung and Ko, Young Gun and Kim, Jung-Gu and Hamad, Kotiba , title =

    Jaafreh, Russlan and Pereznieto, Santiago and Jeong, Seonghun and Widiantara, I. Putu and Oh, Jeong Moo and Kang, Jee-Hyun and Mun, Junyoung and Ko, Young Gun and Kim, Jung-Gu and Hamad, Kotiba , title =. Int. J. Energy Res. , year =

  22. [30]

    Digital Discovery , year =

    Therrien, Félix and Abou Haibeh, Jamal and Sharma, Divya and Hendley, Rhiannon and Mungai, Leah Wairimu and Sun, Sun and Tchagang, Alain and Su, Jiang and Huberman, Samuel and Bengio, Yoshua and Guo, Hongyu and Hernández-García, Alex and Shin, Homin , title =. Digital Discover...

  23. [31]

    and Yang, Qing and Cubuk, Ekin D

    Sendek, Austin D. and Yang, Qing and Cubuk, Ekin D. and Duerloo, Karel-Alexander N. and Cui, Yi and Reed, Evan J. , title =. Energy Environ. Sci. , year =

  24. [32]

    Jalem, Randy and Kanamori, Katsumi and Takeuchi, Ichiro and Nakayama, Masanobu and Yamasaki, Hiroyuki and Saito, Takuya , title =. Sci. Rep. , year =

  25. [33]

    He, Xingfeng and Zhu, Yizhou and Mo, Yifei , title =. Nat. Commun. , year =

  26. [34]

    arXiv , year =

    Yang, Han and Hu, Chenxi and Zhou, Yichi and Liu, Xixian and Shi, Yu and Li, Jielan and Li, Guanzhi and Chen, Zekun and Chen, Shuizhou and Zeni, Claudio and Horton, Matthew and Pinsler, Robert and Fowler, Andrew and Zügner, Daniel and Xie, Tian and Smith, Jake and Sun, Lixin a...

  27. [35]

    Loew, Antoine and Sun, Dewen and Wang, Hai-Chen and Botti, Silvana and Marques, Miguel A. L. , title =. npj Comput. Mater. , year =

  28. [36]

    and Ceder, Gerbrand , title =

    Deng, Bowen and Zhong, Peichen and Jun, KyuJung and Riebesell, Janosh and Han, Kevin and Bartel, Christopher J. and Ceder, Gerbrand , title =. npj Comput. Mater. , year =

  29. [37]

    ACS Mater

    Du, Hongwei and Huang, Xiang and Hui, Jian and Zhang, Lanting and Zhou, Yuanxun and Wang, Hong , title =. ACS Mater. Lett. , year =

  30. [38]

    Parlinski, Krzysztof and Li, Z. Q. and Kawazoe, Yoshiyuki , title =. Phys. Rev. Lett. , year =

  31. [39]

    Togo, Atsushi , title =. J. Phys. Soc. Jpn. , year =

  32. [40]

    Togo, Atsushi and Chaput, Laurent and Tadano, Terumasa and Tanaka, Isao , title =. J. Phys.: Condens. Matter , year =

  33. [41]

    , title =

    Jain, Anubhav and Ong, Shyue Ping and Hautier, Geoffroy and Chen, Wei and Richards, William Davidson and Dacek, Stephen and Cholia, Shreyas and Gunter, Dan and Skinner, David and Ceder, Gerbrand and Persson, Kristin A. , title =. APL Mater. , year =

  34. [42]

    and Persson, Kristin A

    Ong, Shyue Ping and Richards, William Davidson and Jain, Anubhav and Hautier, Geoffroy and Kocher, Michael and Cholia, Shreyas and Gunter, Dan and Chevrier, Vincent L. and Persson, Kristin A. and Ceder, Gerbrand , title =. Comput. Mater. Sci. , year =

  35. [43]

    and Christensen, Rune and Dułak, Marcin and Friis, Jesper and Groves, Michael N

    Larsen, Ask Hjorth and Mortensen, Jens Jørgen and Blomqvist, Jakob and Castelli, Ivano E. and Christensen, Rune and Dułak, Marcin and Friis, Jesper and Groves, Michael N. and Hammer, Bjørk and Hargus, Cory and Hermes, Eric D. and Jennings, Paul C. and Jensen, Peter Bjerre and ...

  36. [44]

    Aitchison, John , title =. J. R. Stat. Soc. B , year =

  37. [45]

    , title =

    Kolouri, Soheil and Zou, Yang and Rohde, Gustavo K. , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , year =

  38. [46]

    IEEE Trans

    Bachoc, François and Gamboa, Fabrice and Loubes, Jean-Michel and Venet, Nil , title =. IEEE Trans. Inf. Theory , year =

  39. [47]

    and Schölkopf, Bernhard , title =

    Gretton, Arthur and Bousquet, Olivier and Smola, Alexander J. and Schölkopf, Bernhard , title =. Algorithmic Learning Theory , year =

  40. [48]

    Efron, Bradley , title =. Ann. Stat. , year =

  41. [49]

    Benjamini, Yoav and Hochberg, Yosef , title =. J. R. Stat. Soc. B , year =

  42. [50]

    Kraskov, Alexander and Stögbauer, Harald and Grassberger, Peter , title =. Phys. Rev. E , year =

  43. [51]

    and Rizzo, Maria L

    Székely, Gábor J. and Rizzo, Maria L. and Bakirov, Nail K. , title =. Ann. Stat. , year =

  44. [52]

    Chen, Tianqi and Guestrin, Carlos , title =. Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discovery Data Mining , year =

  45. [53]

    and Kennard, Robert W

    Hoerl, Arthur E. and Kennard, Robert W. , title =. Technometrics , year =

  46. [54]

    ACS Appl

    Lu, Zhuole and Adeli, Parvin and Yim, Chae-Ho and Mercier, Patrick and Abu-Lebdeh, Yaser and Singh, Chandra Veer , title =. ACS Appl. Energy Mater. , year =

  47. [55]

    and Cubuk, Ekin D

    Sendek, Austin D. and Cubuk, Ekin D. and Antoniuk, Evan R. and Cheon, Gowoon and Cui, Yi and Reed, Evan J. , title =. Chem. Mater. , year =

  48. [56]

    Aghoghovbia, Ogheneyoma and Rurali, Riccardo and Al-Fahdi, Mohammed and Hu, Ming , title =. Chem. Mater. , year =

  49. [57]

    2025 , publisher =

    Togo, Atsushi , title =. 2025 , publisher =. doi:10.48505/nims.4197 , note =

    2025 doi

  50. [58]

    Journal of Physics: Condensed Matter , year =

    Togo, Atsushi and Chaput, Laurent and Tadano, Terumasa and Tanaka, Isao , title =. Journal of Physics: Condensed Matter , year =

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.